合理的粒子群算法( Particle Swarm Optimization Algorithm, PSO) 的参数设置, 可以提高算法的优化效率、避免陷入局部最优值, 但常用参数设置对于特定优化问题, 如新安江模型模拟, 不具普适性。为分析种群规模 pop 、惯性权重 w 、学习因子 c1 和 c2 以及速度位置相关系数 m 这 5 个粒子群参数对新安江模型模拟结果的影响, 对每个参数取 5 个不同水平, 应用 L25 ( 56 ) 正交表, 设计了正交试验。通过对试验结果进行分析, 得出了参数对 PSO 算法性能的影响能力和最优的参数组合方案(pop = 80, w = 1.3~ 0.4 线性递减, c1 = 1.85, c2 = 2.5, m = 0.05) 。通过 极差分析和方差分析, 得出参数 pop 和w 对模型模拟结果具有高显著性, 其他三个参数对模型模拟结果不显著。 将不同 PSO 参数组合应用于新安江模型模拟, 证明了合理的 PSO 算法参数设置可以有效提高新安江模型模拟精度。通过对各因素分别进行趋势分析, 得到了因素取值变化趋势与模型结果变化趋势的相关关系。本文提出的方法为如何寻找某一特定应用情景下的 PSO 算法参数组合提供了一种借鉴。

The reaso nable parameter settings in the particle swarm optimization algorithm canimprove the optimization efficiency and avoidfalling into the local optimum. However, common parameter settings are not universally applicable to specific optimization problems, such as the simulation of Xin'anjiang model. In this study, we conducted orthogonal tests to study the influence of 5 particle swarm parameters on the simulat ionresults of Xin’anjiang model. Through the analysis of the test results, we revealed the influence of parameters on the performance of PSO algorithm and obtained the optimum parameters ( pop = 80, w = linear regression from 1.3 to 0.4, c? = 1. 85, c? = 2. 5, m= 0. 05) . Through range analysis and variance analysis, we found that the parameters pop and warehighly significant to the simulation results, and the other three parameters are not significant to the simulation results. The different PSO parameter sets were applied to Xin’anjiang model simulation, and proved that the reasonable PSO algorithm parameter setting can effectively improvet he simulation accuracy of Xin'anjiang model. Through the trend analysis of each factor, we obtained the relationship between the change trend of the factor value and the change trend of the model result. The method presented in this paper can providereference for finding the parameters of PSO algorithm in a specific application scenario.

“十三五”国家重点研发计划( 2016YFC0402204)